Monday, September 08, 2014

Common genetic variants associated with cognitive performance

This is a follow up to earlier papers by the SSGAC collaboration -- see First GWAS Hits For Cognitive Ability and SNPs and SATS. Effect sizes found are typically ~ 0.3 IQ points. Someone with 50 more good variants (similar to these) than the average person would be about 1 SD above average in IQ.

Note among the authors names like Pinker, Visscher, Plomin, McGue, Deary, etc. Thank god it wasn't the sinister Chinese who got there first! For more on this topic, including the status of the BGI study, see Genetic Architecture of Intelligence (arXiv:1408.3421).
Common genetic variants associated with cognitive performance identified using the proxy-phenotype method (PNAS, doi: 10.1073/pnas.1404623111)

We identify common genetic variants associated with cognitive performance using a two-stage approach, which we call the proxy-phenotype method. First, we conduct a genome-wide association study of educational attainment in a large sample (n = 106,736), which produces a set of 69 education-associated SNPs. Second, using independent samples (n = 24,189), we measure the association of these education-associated SNPs with cognitive performance. Three SNPs (rs1487441, rs7923609, and rs2721173) are significantly associated with cognitive performance after correction for multiple hypothesis testing. In an independent sample of older Americans (n = 8,652), we also show that a polygenic score derived from the education-associated SNPs is associated with memory and absence of dementia. Convergent evidence from a set of bioinformatics analyses implicates four specific genes (KNCMA1, NRXN1, POU2F3, and SCRT). All of these genes are associated with a particular neurotransmitter pathway involved in synaptic plasticity, the main cellular mechanism for learning and memory.

19 comments:

  1. Nifty. Does anyone have a copy of it? Not showing up in the usual places.

    ReplyDelete
  2. Richard Seiter9:18 PM

    You can find it through Steve's link (but how wasn't obvious to me at first). Direct links:
    paper: http://www.pnas.org/content/early/2014/09/05/1404623111.full.pdf
    supplemental material: http://www.pnas.org/content/suppl/2014/09/06/1404623111.DCSupplemental/pnas.1404623111.sapp.pdf

    ReplyDelete
  3. I swear I read over the link several times and C-fed for the strings 'PDF'/'fulltext'/'download' and found zilch. Freaking PNAS website design...

    ReplyDelete
  4. Richard Seiter9:35 PM

    I believe you. I found it through Steve Sailer's post then went back and looked at PNAS to find it hiding under the PDF icon.

    ReplyDelete
  5. jeffhsu39:42 PM

    Great paper. This method should also work the other way, finding more SNPs for educational attainment from cognitive performance data (even with the lower sample size) if using a method that more explicitly deals with the sub-genome wide threshold hits. The genetic correlation should be able to act as some prior right? Although using educational attainment data to find cognitive performance is the more interesting use case.

    ReplyDelete
  6. jeffhsu39:48 PM

    Wow there are lots of authors on this paper that I didn't know were interested in this field. A lot of these names are familiar in the cardiovascular field.

    ReplyDelete
  7. Richard Seiter10:02 PM

    Looking at the three significant SNPS from Table S4 in http://www.pnas.org/content/suppl/2014/09/06/1404623111.DCSupplemental/pnas.1404623111.sapp.pdf
    rs1487441 - coded allele A, std coeff 0.036, 1000 genomes location 6:98553894
    rs7923609 - coded allele A, std coeff -0.034, 1000 genomes location 10:65133822
    rs2721173 - coded allele T, std coeff -0.034, 1000 genomes location 8:145744429

    rs2721173 is odd in that the ancestral allele (C) is associated with higher IQ. Does anyone have any thoughts about this?
    I wonder if the proxy SNPs given in Tables S9 and S11 might be relevant here.

    The coefficients given above correspond (SD=15) to about 0.5 IQ points. It looks like the 0.3 IQ points number comes from the corrected values in Table S7.

    Does it make sense to double these values when thinking about amount of IQ explained since we have two copies?

    You can use the 1000 Genomes locations above to look at worldwide allele frequencies at http://popgen.uchicago.edu/ggv/
    Note the sign of the coefficients when interpreting.

    ReplyDelete
  8. Emil Kirkegaard7:00 AM

    Some of the references in the paper linked above concern the cor of g and height within populations. It seems to be due to correlated genes (e.g. from assortative mating) as opposed to pleiotropy (genes that cause both height and g).

    ReplyDelete
  9. Davide Piffer7:23 AM

    The frequencies of these 3 SNPs are similar across populations (50 populations from ALFRED and 14 from 1KG). Factor loadings are very high (around 0.8). 9 out of 10 (p=0.01) SNPs in Rietvald et al's 2013 paper are derived (unique to humans).Factor scores are highly correlated to IQ, independently of race or ancestral migrations. See my paper for results:http://biorxiv.org/content/early/2014/08/14/008011

    ReplyDelete
  10. Richard Seiter11:38 AM

    Thanks for the pointer. I just looked at Beauchamp 2011 and Marioni 2014. If I understand correctly Beauchamp's standard ACE model gives: "they suggest that common environment and additive genetic effects account for 59% and 31% of the height–intelligence correlation, respectively" and the assortative mating model moves those estimates towards the genetic side. I was unable to either fully interpret table 5 or come away with a sense of which assortative mating parameters were likely to be realistic (I think these are key to the point you are making). Marioni gave an even higher estimate of h^2: "The genetic correlation for height and g was 0.28 (SE 0.09). Bivariate heritability estimates indicated that the majority (71 %) of the phenotypic correlation between height and g was explained by common additive genetic variants."



    What do you think of Deary's 'system integrity' hypothesis? (thanks for the Deary 2012 PDF on your site)

    ReplyDelete
  11. Bibibibibib Blubb1:33 PM

    Shouldn't it have replicated the previous alleles too?

    ReplyDelete
  12. Bibibibibib Blubb2:30 PM

    Strange Those Africans have a higher overall score than Chinese Dai. Pretty close to Han too.

    ReplyDelete
  13. Steve, did anything ever come of that whole 'restricting genetic testing' thing in China?
    http://www.irdirc.org/?p=2733

    ReplyDelete
  14. BlackRoseML5:33 PM

    I think schizophrenia hits would be a good phenotype for the proxy phenotype method. The "unaffected" relatives of schizophrenia demonstrate impaired working memory and prodromal schizophrenics have lower IQ before they are diagnosed. This means that low IQ can affect the development of schizophrenia (by impairing executive function) or has some correlation it it (indicative global neurological dysfunction).

    ReplyDelete
  15. ZA3845:28 AM

    That paper used rs9320913, rs11584700 and rs4851266. This study reports significance for rs1487441, rs7923609, and rs2721173. The new alleles have quite a different distribution, especially because of rs2721173, where the IQ-boosting allele is much more common among Africans, then Americans, then Europeans and Asians. The average of the new 3 alleles puts African, American, and Asian 1000 genomes populations at par with each other, and moderately below European populations.

    Including them should seriously attenuate the reported correlations with Lynn's national IQs.

    ReplyDelete
    Replies
    1. A friend wrote:

      Nah...only 1 of the 3 SNPs has genome-wide significance (the threshold for it is <10^-8), the other 2 have only suggestive significance (<10^-6). The one with the higher significance (which is probably the only SNP of the 3 that will be replicated) is actually at lower frequencies among Africans as you can see from1KG http://www.ensembl.org/Homo_sapiens/Variation/Population?db=core;r=6:98105518-98106518;v=rs1487441;vdb=variation;vf=1217000

      I will not bother replying to them as it's a waste of time.

      Delete
  16. Endre Bakken Stovner6:20 AM

    Using all 7 SNPs gives a correlation with .855 with Lynn's IQ estimates #awkward

    ReplyDelete
  17. Bibibibibib Blubb3:56 PM

    I read about Somantic Masaicism and that shows that healthy individuals have lots of new mutations, increasingly throughout their lives, even different genomes per cell. Its prevailent in the human brain.

    http://www.ncbi.nlm.nih.gov/pubmed/21496937
    http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3018723/
    http://www.cell.com/ajhg/abstract/S0002-9297(14)00312-7

    Would this effect the study or does it already account for it?

    ReplyDelete
  18. Emil Kirkegaard5:57 PM

    Jensen wrote in his 1998 book that the height x g cor is not due to pleiotropy because it is not found within siblings. See The g Factor, p. 146. http://arthurjensen.net/?page_id=9 The genetic correlation you mention, could that be correlated genes instead of pleiotropy? I don't see how pleiotropy is possible if the within family cor is not there.

    Plausible. We need full genome sequencing to properly test it, if the problems are caused by rare genes not covered by standard SNP sequencing. Per the evolution of the price so far, it can't be far away in time. :) https://www.genome.gov/sequencingcosts/

    ReplyDelete